From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Brownell Subject: Re: drivers/usb/musb/musb_io.h Date: Fri, 15 Aug 2008 19:05:31 -0700 Message-ID: <200808151905.32383.david-b@pacbell.net> References: <20080814215200.27f79a59.akpm@linux-foundation.org> <20080815222238.GC25069@frodo> <200808151853.08302.david-b@pacbell.net> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from smtp123.sbc.mail.sp1.yahoo.com ([69.147.64.96]:22943 "HELO smtp123.sbc.mail.sp1.yahoo.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1750928AbYHPCFe (ORCPT ); Fri, 15 Aug 2008 22:05:34 -0400 In-Reply-To: <200808151853.08302.david-b@pacbell.net> Content-Disposition: inline Sender: linux-arch-owner@vger.kernel.org List-ID: To: me@felipebalbi.com Cc: Felipe Balbi , Andrew Morton , linux-usb@vger.kernel.org, linux-arch@vger.kernel.org, Paul Mundt On Friday 15 August 2008, David Brownell wrote: > Someone commented that this is what caused creation of > the mmio_insl() family of routines ... > > ... but I notice that at least on ARM (current primary home > for this driver!) those calls don't end up using optimized > code. Since those optimimzed loops are a *big* win in terms > of performance -- using LDM/STM to burst memory acccess to/from > registers -- I'd rather not go that way for now. Oh, and worth remembering: *WHEN* a clean solution for this is available, it should be usable in the NAND code. And surely in other places which sometimes need PIO access to a FIFO... I was surprised to see that 16% improvement in read performance with an 8-bit NAND device, just by switching from code looking like generic mmio_insb() over to __raw_readsb(). Better IMO to have the faster code be what the NAND core uses by default, than to need to modify every platform's NAND driver to do that. - Dave